Often, constructing humanlike artificial intelligence starts with deconstructing humans. Take fingerprints, for instance.
When washing the dishes, we intuitively adjust our grip on each soapy plate, cup and spoon while taking into account our fingerprint structure. It just doesn't cross our mind because we chalk it up to reflex -- and for the longest time, so did scientists. No one had any equations to unravel how this works because, well, it didn't quite matter. It just worked. But the rise of robotics has complicated things.
For a robot to do this, we have to figure out precisely what's going on, and even turn that knowledge into legible code. Decoding fingerprint grip matters now, and researchers are finally trying to find a new law of physics to explain it.
In a sense, physical knowledge and the code-ability of human traits are prerequisites for robotic programming. And that introduces an important question for the future of lifelike AI: Are there aspects of human consciousness that will never meet these criteria?
It appears that, per some philosophers, there may be. And after reading through two absolutely mind-melting thought experiments, you might agree. Or you might not.
A woman named Mary lives in a little house. She has never left. When she looks around her abode, and out its windows, everything appears black, white or some variety of gray. Mary can't see color, but she often wonders, "What do those people on my black-and-white TV mean when they speak of red roses?"
Suppose Mary's room holds a magic library. This hypothetical place contains books with every snippet of information about the color red. And I mean everything. To quench her thirst for knowledge, Mary reads all of it.
She learns about red electromagnetic wavelengths, how crimson makes people feel, the clearest descriptions of scarlet, analogies about cherries and anything else you can think of. Plus more. No one knows more about red than Mary. Then, she finishes her reading… and decides to step outside her house.
To Mary's surprise, she sees color. She was never colorblind. Her house, furniture and electronics were merely built in black and white, and her windows filtered the outside world in monochrome.
Then something big happens. Mary sees a red apple, the color of her expertise. Her jaw drops. She learns something new about red. But... that's weird. Why wasn't this knowledge somewhere in her library? It had everything one could possibly learn about the color red, right?
This story is a take on the famous thought experiment from 1986, "What Mary Didn't Know," by philosopher Frank Jackson, and the intangible piece of knowledge about red that Mary just gathered is called qualia.
Simply put, qualia defines knowledge only attainable through conscious experience.
It's like the subjective information you received the first time you heard your favorite song. You might've felt shivers down your spine and told your friends, "You have to listen to it to understand." Magically blasting their brains with the depths of music theory and acoustic science probably wouldn't do the trick. Until they hear it, they wouldn't know the song like you do.
Qualia might be why even the best neuroscientists, psychologists and poets likely can't explain the pain of heartbreak well enough to make someone who has never experienced heartbreak truly understand.
And, back to Mary, all the world's physical information about red wasn't sufficient to teach her what looking at the hue would really be like. "When she is let out of the black-and-white room or given a color television, she will learn what it is like to see something red, say," Jackson writes. "This is rightly described as learning -- she will not say 'ho, hum.'"
Though the theory has been tossed around for years, it's still a pretty popular argument for how some knowledge is indescribable by language and unique to human consciousness. That means, if qualia is a real force, its inner workings would be incredibly hard to write down, and therefore, program. It could well be a barrier between humans and AI.
On the other hand, maybe not. Maybe we can decode it somehow, the way we're slowly learning about fingerprint grip dynamics.
Short answer: We don't know. Experts have argued in both directions, and some have come up with fresh angles. But most are stuck behind hypothetical walls, and the fact of the matter is that qualia has no scientific explanation.
OK, from what I've told you about Mary, you've probably thought of some objections to qualia. You wouldn't be alone: Thought experiments are often studded with loopholes, and Mary's room is no exception.
Some counterarguments put forth that shadows in the room could've carried color pigments. Others say the "magic library" would've given Mary knowledge in a way we can't conceive. To the latter point, a fascinating refutation -- and one strikingly relevant to our big question about AI -- comes from philosopher Daniel Dennett.
In a nutshell, Dennett suggests that if Mary really had all the information about red, wouldn't she be sort of omniscient? She wouldn't merely know about the color like a standard human. Theoretically, she would've learned about red "qualia," if it so exists, as part of her red literature, right? And from this line of thinking, we might extrapolate that qualia is indeed poised to level up AI, but we just haven't figured out how to harness it yet.
Well, maybe, but this feels like a dead end. We can't know for sure whether Mary would have such powers. We're not omniscient, so we don't even know what those powers would look like. Thus, Dennett says, let's forget about Mary being human to remove these constraints.
"Thinking in terms of robots is a useful exercise, since it removes the excuse that we don't yet know enough about brains to say just what is going on that might be relevant, permitting a sort of woolly romanticism about the mysterious powers of brains to cloud our judgment," Dennett writes.
Welcome to thought experiment number two.
RoboMary is an iteration of a class of bots called Mark 19, but unfortunately, was built without color vision and is waiting for an upgrade. Until then, RoboMary's "eyes," or video cameras, transmit information in only black and white.
"RoboMary's black-and-white cameras stand in nicely for the isolation of human Mary, and we can let her wander at will through the psychophysics and neuroscience journals reading with her black-and-white-camera eyes," Dennett writes.
Basically, she's looking through her own version of the magic library. But RoboMary takes things a step further.
She learns how Mark 19 color inputs work, then, "using her vast knowledge, she writes some code that enables her to colorize the input from her black-and-white cameras," Dennett writes. She modifies herself in a way human Mary can't.
This new setting allows her to look at an apple, for instance, with her black-and-white vision, then accurately imagine it as the correct color code for Mark 19 bots. RoboMary begins automatically applying the setting to everything as she explores the world. But here's where she really sets herself apart. She observes other working Mark 19s, dissects how they react to various colors and adjusts herself accordingly.
At this point, RoboMary knows what every color input is and reacts to them the exact same way as any other Mark 19.
The big day arrives. RoboMary's color sensors are fired up.
"When she finally gets her color cameras installed, and disables her colorizing software, and opens her eyes, she notices… nothing. In fact, she has to check to make sure she has the color cameras installed," Dennett writes. "She already knew exactly what it would be like for her to see colors just the way other Mark 19s do."
I get chills thinking about this. By changing her settings, RoboMary appears to have simulated qualia for herself. But I also can't stop imagining a much scarier situation.
What if RoboMary opened her eyes… and everything was different?
The Mary saga doesn't stop here.
Notwithstanding tons of other modifications – some of which come from Jackson himself to refine the original argument – Dennett's entire work is incredibly meticulous, too.
He addresses countless objections you might feel as you think about Mary, and RoboMary, and later into a complex scenario that doesn't allow RoboMary to tamper with her settings at all, to see whether qualia is still preserved. There's even a follow-up to Dennett's article entitled "What RoboDennett Still Doesn't Know."
But as with all philosophical thought experiments, the purpose of Mary and RoboMary isn't to tell you a truth. It's to force you to think about the options and find truth on your own.
A few I've come up with are these: Perhaps AI must be built just like RoboMary to gain qualia. Or perhaps robots can be programmed as conscious in a broader sense – that is, if we can find a way to mathematically explain consciousness as a whole. A "conscious" robot could potentially explore the world as we do, and therefore gain qualia like us.
Or maybe qualia isn't what we think it is. Jackson's story makes a convincing argument that once Mary looks at red, something definitely happens. It's been given the name "qualia" and attributed to learning something new, but what if it's a combination of many things with many names and has nothing to do with learning?
Or… maybe, just maybe, qualia really is an untouchable, unprogrammable barrier between human consciousness and AI.
These angles are just the tip of the iceberg, and could (probably will) be disproven as the years go by, if not already. But recall the reason RoboMary was invoked in the first place: to imagine an entity that surpasses human limitations.
It's a far-fetched thought experiment because, as humans, we're restricted. All we can do is speculate.